Daedalab – AI agent safety proxy with real-time permission control

Summary: Daedalab is a proxy that intercepts AI agent tool calls to block dangerous actions before execution, providing real-time classification and approval via Telegram. It enforces hard budget limits and logs all activity for audit purposes, enhancing control over AI-driven operations.

What it does

Daedalab sits between AI agents and LLM providers, classifying tool calls deterministically by reading function names from JSON and blocking critical actions like file deletion or forced git pushes. It requires a single environment variable change and supports multiple frameworks including Claude Code, OpenClaw, LangChain, CrewAI, and AutoGen.

Who it's for

It is designed for developers and teams running AI agents who need to prevent costly or destructive automated actions in production environments.

Why it matters

It prevents unintended destructive operations and excessive spending by enforcing permission gates and budget caps, providing audit trails and manual approval workflows.